Valoración del producto

Considering the state of science reviewed in the literature, this research study was designed to investigate factors which influence SPD among nurses working in Canadian hospital settings.

While job satisfaction of Canadian nurses has been studied, the approach developed for this dissertation to assess SPD offers an alternative construct which removes the noise of extrinsic personal factors which limited interpretation of previous studies. A regression model is built to predict SPD by use of factors identified in the review. Based on the best available evidence, it was hypothesized that the factors daily distress (DD), organizational culture (OC), unit support (US), leadership (L), quality rating (QR) and professional equity (PE) would affect the SP of Canadian hospital nurses. In addition, a structural equation model (SEM) was built to further investigate the relation among SPD and job factors and to confirm or support the regression model. The SEM developed incorporated dimensions of factors to examine their specific effects on SPD.

19 2.3 Summary

In this chapter factors affecting SPD of nurses were identified by use of available evidence and included: DD, OC, US, L, QR and PE. Missing links in our current understanding of these factors affecting SPD of nurses and the need for advanced analytical methods such as SEM to investigate complex models were incorporated in this study. More details of analysis and methods to test model developed are detailed in the next chapter.

20 Chapter 3: Methods

Preface

The study design including rationale for the location, determinations of sample setting, frame and size are described in this chapter. In addition, data collection procedures are described in detail and methods used to measure selected variables are provided along with their

psychometric properties. Data management procedures are discussed to provide information related to data integrity and missing values. Ethical approval processes and limitations in collecting data including mitigating techniques used to avoid biases are reviewed.

3.1 Project description and data collection

A twin-site, full census design was used to test the hypothesized model. This model was specifically chosen to identify factors affecting satisfaction with performance of duties among nurses in Canada by studying two university-based health regions of similar size within Canada – Saskatoon, Saskatchewan and Halifax, Nova Scotia. Regression modeling and structural

equation modeling were used in this study. All nurses in the heart and stroke units within the two health regions were invited to participate. Data from all nurses who actually participated from the two units at the two health regions were used in the regression and structural equation modeling.

3.2. Sample Size calculations for the proposed analysis for modeling:

Regression modeling

Sample size calculations for regression modeling is done based on the two common rules of thump suggested by Green in 1991 as cited by Tabachnick & Fidell, 2014. Sample sizes for regression modeling are dependent on a number of factors including: ”desired power”,” alpha level”,” number of predictors” and expected “effect sizes” (Tabachnick & Fidell, 2014). The guidelines used in regression modeling were: assumption of a medium-size relationship between independent and dependent variables, alpha value of 0.05, and Beta value of 0.20 (Green, 1991 as cited by Tabachnick and Fidell, 2014). Based on six independent variables (IVs) (i.e.

predictors) to explain the dependent variable (DV) (i.e. Satisfaction with Performance of duties (SPD), the number of cases (n) is determined by use of equation 3.1, where m is equal to the number of IVs to be included in the regression model.

21

50 + 8m = n………eq. 3.1 Therefore;

50 + (8*6) = n 98 = n

When testing for correlations among a number of IVs, the number of cases is determined by equation 3.2.

104 + m = n……….eq. 3.2 Therefore;

104+6 = n 110 = n

Structural equation modeling (SEM)

The sample size for SEM is determined by use of empirical rules of measures recommended by various SEM researchers. Marsh and Yeung (1997) recommend four items per construct each of which have local (specific latent variable) and global (overall model) aspects. Nicolaou and Masoner (2013) suggested that a minimum absolute sample size is a practical necessity since underlying estimation theory is asymptotic. Lomax and Schumacher (2004) and, Boomsma and Hoogland (2001) recommend 100 observations and Hu and Bentler (1999) recommend a

minimum of 250 observations. However, Bentler and Yuan (1999) identified new testing

techniques specifically for SEM that produce sufficiently robust model estimates for as few as 60 subjects, as cited in Tabachnick and Fidell (2014). Similarly, two recent simulation studies demonstrated small sample sizes can be sufficient for SEM. Wolf, Harrington, Clark and Miller (2013) found sample size requirements of 30 and higher for simple Confirmatory Factor

Analysis (CFA) with four indicators loadings of approximately 0.80. Sideridis, Simos,

Papanicolaou and Fletcher (2014) found that sample sizes of 50-70 would be enough for a model to keep low Type-I error rates.

The N:q rule based on maximum likelihood estimation (ML) is a popular rule for determining sample size and model complexity and is the default method in most SEM computer programs (Jackson, 2003). “In ML estimation, Jackson (2003) suggested that researchers think about

22

minimum sample size in terms of the ratio of cases (N) to the number of model parameters that require statistical estimates (q)”. The optimal sample size to parameters ratio recommended is 20:1, but a minimum of 10:1 is acceptable (Jackson, 2003). When the N:q ratio decreases to a value below 10:1, the trustworthiness of the results also approach to a minimal level. The online calculator provided by Soper (2017) was also used to calculate sample size assuming an effect size of 0.15, desired statistical power level of 0.8, number of latent variables 1 and number of observed variables 6 with a probability level of 0.05. The Soper’s calculator estimated a recommended minimum sample size of 200.

Based on reviewing many field experiences Kline (2005) concluded that a minimum sample size of 200 is required for valid structural equation modeling. Subsequently, Hox, Maas and

Brinkuis (2010) suggested that a sample size of 50 is sufficient when the investigation is primarily to seek factor loadings. Based on these approaches for determining sample size and considering the practical realities of the scope of this twin-site study, a combined sample size of 200 for the two sites was deemed to be adequate.

Exploratory Factor Analysis (EFA)

A good general rule of thumb for factor analysis is provided by Tabachnick and Fidell (2014) is 300 cases or a more a subject to item ratio of 10:1 suggested by Costello and Osborne 2005.

Comrey and Lee (1992) as cited in Williams, Onsman and Brown (2010) provided the following guide for sample sizes “– 50 as very poor, 100 as poor, 200 as fair, 300 as good and 500 as very good and 1000 as excellent”. However, Gaudagnoli and Velicer (1998) and Velicer and Fava (1998) shown that solutions with several high loading marker variables (greater than 0.8) do not require as many cases (Velicer & Fava, 1998). Considering the rules of thumb above, it has determined that our combined sample size of 234 for both sites provided adequate sample size for the factor analysis.

3.3 Location of study and sample

Data were collected from nurses of the cardiology and stroke units of both Saskatoon and Halifax health regions, two medium sized urban university teaching hospitals. Upon receiving ethics approval from the University of Saskatchewan and Saskatoon Health Region, approval was also granted by the Capital District Health Authority in Halifax (Appendix A and B).

Discussions were held with the division leaders of the cardiology and neurology departments of

23

both health regions. The division leadership included physicians, nurses and inter-professionals (the collective term used to represent physiotherapists, occupational therapists, social workers and other therapists). A champion was identified at both sites to review instruments and conduct surveys. Study coordinators were hired at both sites and worked with two co-principal

investigators (Co-PIs) and champions selected at each site. Several presentations were made by the Co-PIs to the night and day cardiology and neurology physicians and staff at each site.

Separate survey packages were prepared for the cardiology and neurology (stroke) departments at each hospital (Appendix E and F). Each package contained a description of the study and associated objectives, an invitation to participate (Appendix A and B) that conformed to

requirements as established by Tri-council Research Ethics, and the questionnaire itself. Nuances in questionnaire phrasing were determined by health region therefore terminology was slightly different between the questionnaires provided to the two hospitals. Separate questionnaires were developed for physicians, nurses and inter-professional staff from cardiology and neurology based on the Lepnurm et al. (2012) study of nurses and physicians in cardiology and neurology departments of the Saskatoon Health Region in 2009. Refinements of the original questionnaire were made for this study which was carried out in 2015.

Study champions piloted the questionnaire with small groups of nurses from their respective units to ensure that items were understandable and terminology appropriate. Prior to piloting the questionnaire with their own staff, the champions met with study coordinators and Co-PIs on several occasions in person followed by clarifications by e-mail and telephone to further develop draft questionnaires until everyone was satisfied.

The instruments were distributed on patient care units of cardiology and neurology at both sites.

Most participants found time to complete the questionnaire during their respective shifts when questionnaire was received from the study coordinator. Some took their questionnaire home and then returned it completed on their subsequent shift. Participants returning completed

questionnaires to the collection box placed in their unit received a small present to recognize their time and effort and were well received. The study coordinator worked with the nursing leaders of each unit to devise three questionnaire completion blitzes in conformity with the approach outlined by Dillman (2002; 2011) and described in the Nursing Working Environment Study by Lepnurm et al. (2012) such as;

24

 Include all staff documented in the respective units working half-time or more;

 Study coordinators at each site should act as custodians of the staff lists and assign each questionnaire a code number to ensure confidentiality of data;

 Questionnaires should be distributed by the study coordinators and returned to a collection box in the respective unit to protect privacy of participants, minimize interruptions of clinical duties and to be cost efficient.

 Immediate distribution of participation gifts (travel mugs) to participants by use of the honour system (by the third blitz the travel mugs were placed by the collection box).

 Study coordinators checked questionnaire collection boxes (Stanley Tool Boxes with a slot cut in the top, locked by padlock) every second day throughout the duration of the study, February to May, 2015).

3.4 Criteria for inclusion and exclusion of study participants

Study subjects for this research project consisted of nursing staff of the cardiology and stroke units of hospitals in the Saskatoon and Halifax health regions. All Registered Nurses (RN) and Licensed Practical Nurses (LPN) working at least half-time, for a minimum of one year,

delivering patient care on the cardiology and stroke units were recruited to participate in study.

Nurses were excluded if they were on: vacation, maternity leave, leave of absence, or suffering from an illness. Nurses with significant administrative duties were also excluded.

Overall, 371 nurses, 168 from Halifax and 203 from Saskatoon were included in the study and consisted of 317 RNs and 54 LPNs. In Halifax, 119 nurses returned completed questionnaires, for a response rate of 70.83%, and in Saskatoon, 117 nurses returned completed questionnaires for a response rate of (57.64%). In cardiology 128 nurses (59.53%) and 108 nurses in neurology (69.23%) returned completed questionnaires (Table 3-1). The overall response rate was 63.6%

(62.77% for RNs and 68.51% for LPNs) (Table 3-1).

25 Table 3-1. Response rate of nurses by region and unit.

Region Unit Nurse

The study sites were the health regions of Saskatoon and Halifax.

3.5.1 Saskatoon health region (SHR) (Sourced from Wikipedia)

At the time of the study, the SHR was the largest health region in the province of Saskatchewan and is the location of the only medical and largest nursing school. This region services

26

approximately 360,000 local residents in more than 100 cities, towns, villages, Regional Municipalities, and First Nation communities. The SHR is also a provincial referral centre, providing specialized care to people from across the province. SHR is an organization which provides services and programs to more than 70 facilities of which ten are hospitals (including three tertiary hospitals in Saskatoon), 29 are long term care facilities, and numerous primary health care sites, public health centres, mental health and addictions centres (Wikipedia SHR) Note: The regional health system is currently transition to form a single provincial health authority and officially launched on December 4, 2017.

3.5.1.1 Cardiology and stroke units

Medical and nursing care for 80% of heart attack or stroke victims within the SHR are provided by four units at the Royal University Hospital and the Rehabilitation Unit of the City Hospital.

Initial admission is usually through the respective Emergency Departments, followed by transfer to the Coronary Care Unit, Acute Stroke Unit or Cardiac Care Unit. Smaller proportions of heart attack and stroke patients with less severe conditions are diagnosed and treated at St. Paul’s Hospital. This study involved nursing staff from the Coronary Care, Cardiac Patient Care and Acute Stroke Care Units of Royal University Hospital. The catchment population covers the northern half of Saskatchewan (approximately 600,000 people).

3.5.2 Capital district health authority (CDHA) (Sourced from Wikipedia)

Out of a total of 9 health authorities the CDHA was the largest in the province of Nova Scotia. In 2015 it was integrated into the province-wide Nova Scotia Health Authority. CDHA was

delivering essential health services in the Halifax Regional Municipality and in the Municipality of the District of West Hants, for a combined population of over 400,000 residents. This is approximately 42% of the provincial population of 953,869 (Statistics Canada 2018) 40% of the provincial population. The CDHA was also responsible for the advanced level specialized acute care to residents throughout Atlantic Canada as the largest teaching hospitals affiliated the Dalhousie university medical school are located within this health authority. Nearly 10,000 staff are employed within CDHA (Wikipedia CDHA).

3.5.2.1 Cardiology and stroke units

Medical and nursing care for 90% of heart attack or stroke victims within the CDHA is provided

27

at the Halifax Infirmary Site of the Queen Elizabeth Hospital. After initial admission and diagnosis in the Emergency Department, treatment for heart attack victims is provided in the Intermediate Care, Coronary Care or Cardiac Care Units. For stroke victims, treatment is provided in the Acute Stroke Unit and some patients continue to receive care on the Stroke Unit at the Nova Scotia Rehabilitation Centre. A small fraction of heart attack and stroke patients, those with less severe conditions are diagnosed and treated within the general medical floors at Dartmouth General Hospital. This study focused on heart attack and stroke care provided within the Halifax Infirmary where the vast majority of cases are treated. The catchment population covers the provinces of Nova Scotia and Prince Edward Island and part of New Brunswick (approximately 1,200,000 people).

3.6 Instrumentation

The Lepnurm et al. (2012) nursing questionnaire used in the Saskatoon Nursing Work

Environment Study of 2009 was updated and refined for use in this twin site study. The process of refinement involved two steps; first the content of each of the dependent and independent variables was reviewed by the research team in collaboration with nursing managers at both sites; and second, the phrasing of each item was adjusted to use the appropriate clinical

terminology for cardiology and neurology units and terminology used at the two health regions.

The questionnaires were reviewed by researchers of the MERCURi Research Group (A university of Saskatchewan based researchers from multiple faculties including public health, pharmacy and nursing) and funded by Canadian Institutes of Health Research (CIHR). The two nursing faculty on the research team assumed the greatest responsibility for conceptual content while psychometric experts ensured that concepts were properly measured. Similar variable structure to 2009 was used in 2015 since the relationships among the variables used in 2009 were strong (Lepnurm et al., 2012). The survey consisted of various measures prepared specifically for variables that are determined to be associated with SPD of nurses in Canadian hospital work environments. Satisfaction with Performance of Duties was treated as the dependent variable and all factors associated with SPD were treated as independent variables. The independent variables included’ organizational culture (OC), organizational support for the unit (US), professional equity (PE), daily distress (DD), leadership (L) and quality Rating (QR). Refinements were then made to each of the variables in consultation with nursing leadership at both sites. The

28

refinement process consisted of several rounds of discussions between the researcher and small groups of staff at each site, and within each of the two clinical divisions. The refinement process proceeded variable by variable until the researchers and nursing leaders were satisfied with the questionnaires. The nursing questionnaires are presented in Appendix E and F.

Pre-testing was conducted with two groups of 4 or 5 nurses from each health region from the Cardiology or Neurology Divisions. Overall pre-testing was conducted by use of four groups consisting of four LPNs and 14 RNs. Collected data was entered into Microsoft Excel (2010) © and then uploaded into SPSS. Frequency distributions for each item were reviewed to ensure normality (slightly positive skews were expected). Negatively worded questions were reverse coded to ensure that subsequent factor analyses and reliability testing could be carried out. Basic SPSS programming for the latent constructs consisting of multi-item measures was done at pre-testing and maximum and minimum values of the pilot data set were checked to ensure responses were in line with the selected scale.

Following the pilot test, some refinements were required and the pre-test group of nurses was asked to complete a smaller questionnaire with just the revised items. Asking them to fill out a full second survey was considered but rejected due to their time constraints.

3.7 Establishment of measurement scales

Because all variables were measured by use of psychometric scales and selected from a

previously validated questionnaires designed for a similar population, the use of factor analyses and reliability testing was deemed appropriate (Tabachnick & Fidell, 2014). Exploratory Factor Analysis (EFA) using principal components method of analysis and select factors that have eigenvalues over 1 as our criteria with varimax rotation was used to verify that new items added to existing scales contributed to overall explanation and that factor structures of refined measures were consistent with the original measures. Next, Cronbach’s reliability was used to test internal consistency of factors within each measure. Since some new items were developed for several of the variables, internal consistency was expected to increase for the factors and some new factors were expected to emerge.

The traditional procedure to validate measures is by comparison of results of a newer measure with the results of an older more established measure; however, such studies require respondents to answer a battery of similar items twice and only a few measures can be tested at any one time

29

to avoid respondent fatigue (Lavarkas, 2008, Ben-Nun 2008). A more common procedure is to compare the factor structures and reliability analyses of sub-sets within the same dataset, but caution must be exercised to ensure that sub-populations are of sufficient size. Exploratory Factor Analysis usually considered as a large sample size technique with a sample size of 50 as absolute minimum ( De Winter, Dodou & Wieringa 2009). Comrey and Lee as cited in

Tabachnick and Fidell 2014, considered “sample sizes of 50 as very poor, 100 as poor, 200 as fair, 300 as good, 500 as very good and 1000 as excellent”. For factor analysis, 300 cases are considered sufficient (Tabachnick & Fidell 2014). For assessment of reliability, Cronbach’s alpha values of .800 or more are considered highly reliable, over 0.700 as good and over 0.600 as acceptable for sub-scales. George and Mallery (2003) as cited in Gliem and Gliem (2003)

provide similar cut-offs and “consider values greater than 0.9 as excellent, 0.8 as good, 0.7 as acceptable, 0.6 as questionable, 0.5 as poor, and below 0.5 as unacceptable”. In this study, validation of Cronbach alpha values for sub-scales of all usable respondents (n=234) were compared with sub-populations by Region: SHR (n=115) or CDHA (n=119); clinical condition:

Heart Attack or Myocardial Infarction (MI n=127) and stroke or cerebro-Vascular Accident

Heart Attack or Myocardial Infarction (MI n=127) and stroke or cerebro-Vascular Accident